TLDR;
This video explores the capabilities of OpenAI's GPT-5 in the context of trading strategies. It tests GPT-5's ability to convert indicators into strategies, improve existing strategies, and create new strategies from scratch. While GPT-5 shows promise, particularly in generating new strategies with favorable risk profiles, it's not significantly better than GPT-3 for improving existing strategies. The video emphasizes the importance of user guidance and iterative refinement to achieve optimal results.
- GPT-5 can convert indicators into trading strategies, providing a foundation for further refinement.
- GPT-5 struggles to improve existing strategies effectively, sometimes eliminating profitable trades.
- GPT-5 excels at creating new trading strategies, often with low maximum drawdown, making them suitable for leveraged trading.
- Human guidance and iterative improvement are crucial to maximizing the potential of AI-generated trading strategies.
Intro [0:00]
The video introduces a test of OpenAI's new GPT-5 model for trading strategies, examining its ability to convert indicators into strategies, improve existing strategies, and create entirely new strategies. The presenter emphasizes the importance of understanding the model's strengths and weaknesses to leverage it effectively in trading. He encourages viewers to engage with the content by liking, subscribing, and clicking the bell for notifications.
Better than o3? [1:04]
The presenter questions whether GPT-5 is superior to GPT-3, noting that GPT-5 seems aware of their previous interactions, highlighting its strengths in reasoning, coding, and Pine Script. GPT-5 claims to be good at tweaking Pine Script and converting indicators into weekly strategies, which the presenter intends to verify through testing.
Convert Indicator to Strategy [1:37]
The presenter challenges GPT-5 to convert the Lux Algo AI Super Trend Clustering Oscillator indicator into a trading strategy. He explains the indicator's logic, where blue uptrends suggest buying opportunities and red clouds indicate weakness, suggesting selling. The presenter explores two approaches: allowing the AI to determine buy/sell logic and providing specific buy/sell conditions based on the indicator's signals. He references a previous video on converting TradingView indicators into strategies for viewers seeking more information. The initial attempt, where the AI decided the buy/sell logic, resulted in a poor strategy with a -32% return on Bitcoin since 2018. When the presenter specified to buy on blue dots and sell on red dots, the strategy yielded 200% profit with a 37% max drawdown.
Improve Strategy [6:03]
The presenter tasks GPT-5 with improving an existing Gshian channel strategy by reducing trades in sideways markets. He provides the AI with the strategy code, input screenshots, and Bitcoin chart data, instructing it to aim for higher profit and lower drawdown. Despite initial coding errors that required correction, the AI produced a new version with a lower drawdown (30% to less), fewer trades, and a slight increase in profitable trades. However, the AI failed to eliminate trades in sideways markets as instructed, even removing some profitable trades. The presenter expresses disappointment, finding GPT-5 no better than GPT-3 for this task.
New Strategy [9:44]
The presenter aims to extract hidden, high-quality trading strategies from GPT-5's training data by giving it maximum freedom in creating a new strategy. He provides Bitcoin chart data and specifies a goal of super high net profit with a maximum drawdown below 30%. After multiple iterations to correct coding errors, the AI generated a strategy with a 15.46% max equity drawdown and a 61% profitable trade rate. Although the net profit was not exceptionally high, the low drawdown allows for 2x leverage, resulting in a 30% drawdown with 2,400% profit and a 60% profitable trade rate. The presenter recommends further conversation with the AI to improve the strategy and refers viewers to a video on improving trading strategies.
Conclusion [12:53]
The presenter concludes that GPT-5 is not significantly better than GPT-3, suggesting it may be consolidating existing models rather than possessing superior knowledge. He recommends viewers explore pre-existing trading strategies on signum.money and consider the autotrading masterass for advanced portfolio management education.
Automate Strategies [13:39]
The presenter promotes signum.money, highlighting the availability of the Gshian channel strategy for the one-day chart and the Ichimoku strategy for the four-hour chart. He mentions videos demonstrating how to automate these strategies for buying and selling on exchanges.
Become a Master of Automation [13:51]
The presenter promotes the autotrading masterass program for individuals with crypto portfolios of $25,000 or more. He emphasizes the importance of education in building a risk-adjusted portfolio, understanding market cycles, and rebalancing. The program offers limited access and is designed for higher net worth individuals who understand the importance of risk management.